DECCA Repurposed: Detecting transcription inconsistencies without an orthographic standard
C. Anton Rytting, Julie Yelle · 2017
Most language resources and technologies depend on written text, while most endangered languages are primarily spoken.Transcribing speech into text is time consuming and error-prone.We propose a method for finding spelling inconsistencies without recourse to a standard reference dictionary or to a large training corpus, by repurposing a method developed for finding annotation errors.We apply this method to improve quality control of audio transcriptions, with particular focus on under-resourced, primarily oral language varieties, including endangered varieties.